Skip to content
Rohit Suratekar

Selected work — models, papers and code.

Work across fundamental research, genomics, clinical data, AI systems and software. Filter by discipline or by kind of output. The full bibliography and talks are under publications.

Highlights

Clinical & real-world dataModel · 2026

FDA clearance for ECG-AI cardiac amyloidosis detection

FDA 510(k) clearance (K253801)

The cardiac amyloidosis algorithm received FDA 510(k) clearance — the first and only one cleared to detect the disease from a standard 12-lead ECG.

Press release
Read the storyHide the story
Problem
Cardiac amyloidosis is a rare, under-diagnosed heart disease, while a standard 12-lead ECG is routine and widely available.
My role
Led development of the ECG-AI model at nference with Pfizer, then owned the scientific deliverables of the FDA 510(k) submission at Anumana — clinical validation analyses, technical reports and responses to FDA questions.
Result
FDA Breakthrough Device Designation (2023), selection for the FDA TAP pilot, and 510(k) clearance in April 2026 — the first for this indication using a standard 12-lead ECG.

AI & LLM systemsSystem · 2025

Agentic framework for clinical-trial cohorts

~40% faster cohort creation

Specialised agents extract eligibility, match inclusion/exclusion criteria and refine clinical-trial cohorts from EHR and multimodal records, with every decision traced to evidence.

Read the storyHide the story
Problem
Building a clinical-trial cohort from EHR records means reading eligibility criteria and checking patients against them by hand — slow, and hard to reproduce.
My role
Designed the agentic LLM workflow in LangGraph and LangChain — eligibility extraction, inclusion/exclusion matching and cohort refinement, with every decision traced to its evidence.
Result
Cohort creation about 40% faster, with better reproducibility and traceability.

Clinical & real-world dataPaper · 2025

Predicting heart failure from 12-lead ECGs

JACC · AHA 2025 Late-Breaker

AI prediction of heart failure from routine 12-lead ECGs, in a pooled HeartShare/AMP-HF cohort analysis.

Read paper
Read the storyHide the story
Problem
An ECG-AI model for heart failure is only useful if it holds up on people outside the data it was built on.
My role
Led external validation with the HeartShare consortium on pooled research cohorts; reviewed statistical analysis plans and managed relationships with academic partners and CROs.
Result
Published in JACC (2026) and presented as Late-Breaking Science at AHA Scientific Sessions 2025.

All 27 projects

27 of 27